Use Case Description
Building AI factories is a monumental challenge, requiring intensive planning and cross-discipline orchestration to ensure that these complex facilities can be built quickly, without waste, and with optimal use of power and cooling. This type of multi-discipline aggregation and simulation is accelerated with NVIDIA Omniverse™ Libraries and SDKs as well as OpenUSD, unlocking the ability for teams to encode physical and nonphysical data together into a digital twin that developers can leverage inside their applications to run what-if scenarios, space planning, and operational dashboards in real time.
The NVIDIA Omniverse DSX Blueprint is the digital twin manifestation of the DSX reference design, demonstrating to developers how to use Omniverse libraries for design, simulation, and operations across AI factory facilities and their hardware–software ecosystem.
Experience Walkthrough
The application opens to an interactive global viewport. Users will see a “Configurator” panel in the top righthand corner and a left-side toolbar organized into “AI,” “Main,” and “View” sections.
Clicking a site reveals the “Analytics” panel, which shows detailed site information and performance KPIs. Selecting the simulation icon in the left toolbar launches the simulation panel and changes the view to a predefined camera. From here, users can choose between a thermal or electrical simulation. Clicking “Begin Test” initiates the simulation.
During the simulation, a color and/or motion overlay will appear on the 3D assets in the viewport. The “Camera” option in the left toolbar allows users to adjust the viewing angle. The viewport resets once the configuration is finished.
Powered by OpenUSD, the digital twin of this AI factory is built using OpenUSD and SimReady assets, ensuring a standardized, interoperable workflow essential for seamless collaboration, asset reuse, and accurate simulation in complex industrial environments.
Simulation Setup and Execution
To begin, users must configure the data center simulation by setting general, thermal, and electrical outputs in the “Simulations” tab. Users should start with the “Configurator” panel to understand the settings that will be used for both the “Test” and “Simulate” modes.
The configuration settings established in “Configurator” enable two distinct simulation types:
- Test (Electrical Simulation):
- Allows users to simulate various electrical scenarios, such as power failure, using the selected configurations.
- Users select a desired test scenario and click the “Begin Test” button to run it.
- Power Failure Scenario: The "Status" indicates whether the scenario passed or failed. The power system's behavior is visualized in the viewport: green for de-energized, red for energized, and purple for overloaded. Specific performance numbers are available in a table.
- Simulate (Workload and Thermal Ramp-up):
- Allows users to configure and run a workload and thermal ramp-up simulation to gain insight into how workloads affect the data center's thermal and electrical components.
- After selecting “Sim Type” and “Display Type,” users click the “Play” button (either in the “Simulate” panel or at the bottom of the viewport) to start the simulation.
- During the simulation, the colors in the “Workload Distribution” pie chart are displayed directly on the server assets, showing which parts of the server are allocated to each workload.
Architecture Diagram
What’s Included in the Blueprint
The NVIDIA Omniverse DSX Blueprint repo contains:
- Digital twin set of geometry based on the entire DSX reference design for a 50-acre site including compute building and support infrastructure.
- Front-end web application with user interface developed with Omniverse libraries, for interacting with digital twins, viewing simulations, and creating and saving build configurations.
- Simulation-ready assets to accelerate digital twin creation:
- Computational fluid dynamic thermal hot aisle simulation.
- Sample compute configurations for DSX such as GB200 and GB300 NVL72 designs.
- Electrical loading simulation to test various loading configurations.
It is powered by NVIDIA libraries available such as:
- CUDA-X™
- Omniverse Kit
- Warp
- AI Agent NIM™
- RTX™
- Kit-CAE
Key functionalities include:
- Accelerated Time-to-Revenue: Streamline operations through system-level optimization, modular construction, and rapid cluster bring-up.
- Energy-Efficient Performance: Maximize token throughput per power budget while securing early access to power resources.
- Sustainable Resource Management: Minimize environmental impact using dry coolers, heat reuse, and advanced water-usage optimization.
- Intelligent Workload Placement: Coordinate IT and OT systems to automate workload distribution based on real-time power and cooling availability.
- AI-Driven Reliability: Ensure maximum uptime via integrated AI agents and predictive maintenance across IT and infrastructure layers.
- Future-Proof Infrastructure: Design flexible environments capable of supporting multiple hardware generations with minimal downtime.
The NVIDIA Omniverse DSX blueprint provides an end-to-end framework to build, simulate, and optimize gigawatt-scale AI data centers—seamlessly transitioning from custom app development and physically accurate design to high-efficiency, grid-resilient operations.
Value Proposition:
- Build Custom Apps with NVIDIA Omniverse Libraries: Utilize modular OpenUSD libraries to develop specialized digital twin applications. Create a bespoke, extensible foundation tailored to your specific industrial AI and data center workflows.
- Accelerate Development with SimReady Assets: Rapidly assemble environments using standardized, physically accurate SimReady assets. Skip manual modeling and move straight to high-fidelity virtual staging and validation.
- Optimize Design for Power and Thermal Efficiency: Co-design compute, power, and cooling systems in a unified environment. Use simulation to maximize rack density and ensure thermal stability before breaking ground.
- Boost Workload Performance with DSX Max-Q: Scale “Max-Q efficiency” across the data center to deliver up to 30% higher GPU throughput. Turn power constraints into a competitive advantage by maximizing performance-per-watt.
- Streamline Operations with DSX Flex and Exchange: Unify IT and OT into a single “operating system” for your facility. Use AI agents to sync energy demand with grid conditions, ensuring resilient, gigawatt-scale operations.
Omniverse DSX Blueprint is designed to democratize NVIDIA AI factory design and operations. It is for partners who aim to build AI factories efficiently, and at scale by incorporating digital twins into the AI factory life cycle:
- NVIDIA Cloud Providers (NCPs): High-growth cloud entities scaling AI-specialized infrastructure
- AI Labs: Research-driven organizations requiring rapid, high-fidelity environment deployments.
- Sovereign AI: National and regional entities building localized, secure AI infrastructure to ensure data and compute sovereignty.
Omniverse DSX Blueprint serves as the connective tissue for the ecosystem to build digital twins to design and optimize the AI factory lifecycle:
- Colocation and Capacity Providers: Land, power, and shell (LPS) builders and operators—such as Digital Realty, Switch, and Vantage
- Design and Construction (AEC) Firms: Engineering and infrastructure support for LPS owners—such as Jacobs and Bechtel.
- Power and Cooling Solution Providers: Critical hardware and systems infrastructure—such as Vertiv, Schneider Electric, and GE Vernova.
- Channel and Systems Integrators: Partners delivering end-to-end system integration services—such as WWT.
- Independent Software Vendors (ISVs): Providers supporting the design, construction, and ongoing operations of AI factories.
Terms
PUE: Power usage effectiveness. PUE is a metric used to evaluate the energy efficiency of a data center by comparing the total energy consumed by the facility to the energy used by the IT equipment alone.
TCO: Total cost of ownership. TCO is a comprehensive financial metric that assesses the entire lifecycle cost of acquiring, using, and eventually disposing of a product or facility.
Power failure: Interruption or outage of supply power to the AI factory.
CDU failure: Cooling distribution unit failure. This piece of liquid cooling equipment removes heat from the IT equipment.
Chiller failure: Chillers provide cooling capacity for the data center. Chillers control the chilled facility water for CDUs (liquid cooling systems) and CRAH units (air cooling systems).
Workload simulator: Enables the data center designer/operator to run a series of different types of workloads across the AI factory.
Ramp-Up simulator: The power utilization of GPU racks vary between 40% to 100%. The AI factory digital twin thermal simulation shows the impact on the hot aisle air temperature in real time.
FWS temp: Facility water system temperature
SAT temp: Supply air temperature
TCS temp: Technology cooling system
RPP Size: Remote power panel
Terms of Use
Governing Terms: The trial service is governed by the NVIDIA API Trial Terms of Service.
